[Bug ld/20545] New: [avr] Incorrect offsets computed for PC relative jumps with linker relaxation and alignment directives

[saaadhu at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=20545

            Bug ID: 20545
           Summary: [avr] Incorrect offsets computed for PC relative jumps
                    with linker relaxation and alignment directives
           Product: binutils
           Version: 2.27
            Status: NEW
          Severity: normal
          Priority: P2
         Component: ld
          Assignee: unassigned at sourceware dot org
          Reporter: saaadhu at gcc dot gnu.org
  Target Milestone: ---

There are a couple of cases where linker relaxation causes incorrect
computation of addends for relocs, resulting in wrong code or reloc overflow
errors.

$ cat repro.s
        call foo
        nop 
        .p2align        1
        nop
.L618:
        ldi r24,lo8(6)
        brsh .L618
foo:    nop
$ avr-as -mavr5 repro.s -o test.o && avr-ld -mavr5 --relax test.o &&
avr-objdump -S a.out

a.out:     file format elf32-avr


Disassembly of section .text:

00000000 <__ctors_end>:
   0:   03 d0           rcall   .+6             ; 0x8 <__ctors_end+0x8>
   2:   00 00           nop
   4:   00 00           nop
   6:   86 e0           ldi     r24, 0x06       ; 6
   8:   e8 f7           brcc    .-6             ; 0x4 <__ctors_end+0x4>

0000000a <foo>:
        ...

Note that the brsh in the source code jumps to the ldi instruction, whereas the
disassembly of the linker output shows that it jumps to the instruction before
it (nop).

$ cat repro2.s
foo:
        jmp foo
        call foo
.L1:
        brsh .L1
.p2align        1
        nop
$ ~/avr/install/bin/avr-as -mavr5 repro2.s -o test.o &&
~/avr/install/bin/avr-ld -mavr5 --relax test.o && ~/avr/install/bin/avr-objdump
-S a.out

a.out:     file format elf32-avr


Disassembly of section .text:

00000000 <__ctors_end>:
   0:   ff cf           rjmp    .-2             ; 0x0 <__ctors_end>
   2:   fe df           rcall   .-4             ; 0x0 <__ctors_end>
   4:   e8 f7           brcc    .-6             ; 0x0 <__ctors_end>
        ...

Again, the brsh in source code jumps to .L1 (i.e. itself), whereas in the
disassembled output, it jumps to the first instruction (jmp relaxed to rjmp).

-- 
You are receiving this mail because:
You are on the CC list for the bug.